If you read my blog posts or follow me on LinkedIn, you probably know that I’m passionate about the intersection of materials science and sustainability. AI, not so much.
I attended SEMI’s Strategic Materials Conference this month and came away with a new perspective on AI and what it can do.
I still don’t advocate using AI just because we can. Indiscriminate use is a recipe for ballooning greenhouse gas (GHG) emissions associated with the incredible computing power needed to train AI models. In his opening keynote presentation, Bill En of AMD mentioned that, given predicted growth, the energy required for AI will soon equal the amount needed to power all of New York City.
As En said, “To realize the benefits of AI, we must find a way to optimize power without sacrificing performance.” Part of that involves more efficient heterogeneous integration. As Moore’s law reaches its limits, advanced packaging can save the day. Materials also make a difference.
A Materials Revolution
The conference theme was “Materials Revolution: Shaping the Future of Electronics and Beyond, and the Impact of AI.” That’s a mouthful—the point is that AI can shape a materials revolution.
Laura Matz of Merck explained that AI models can shorten the research time for new materials discovery from months to hours. These models can be used to investigate replacements for PFAS molecules, find conductors or dielectrics that reduce on-chip power consumption, or evaluate process gases with lower global warming potential (GWP).
Bringing in AI expands the number of possible candidates to consider. As Matz noted, nearly the entire periodic table is up for grabs for molecule discovery. For initial screening, AI can consider any element that isn’t radioactive. In reality, focusing on a specific application narrows the field.
Consider, for example, materials for a new solid-state electrolyte in batteries that aim to reduce the total lithium content. Ken Kaneko from Microsoft shared a case study done in collaboration with Pacific Northwest National Laboratory (PNNL). Initial screening of over 30 million possible materials resulted in a list of about 500,000 stable compounds. Additional filters, including electrical and physical properties and material cost, narrowed the list to 18 realistic choices. Microsoft and PNNL chose one for synthesis and qualification.
A similar process could be done to select improved materials for any step in front-end or back-end processing. Instead of merely selecting for performance, AI could simultaneously evaluate toxicity, energy requirements, or GWP of inputs and byproducts. Simulations could avoid investing years of research and millions of dollars into regrettable substitutions.
Government Support for Materials Development
The U.S. Department of Commerce is launching an open competition for projects that use AI to advance semiconductor sustainability as part of the CHIPS for America program. The focus is on materials development. They are looking for projects that are university-led and involve industry collaboration.
As speaker Carol Handwerker explained, we need to add environmental impact, human health and safety, and supply chain risk to the typical performance, power consumption, and cost metrics. Optimizing for all these concerns and constraints makes materials discovery much more complex, which is why we need AI solutions. Handwerker advocates using artificial intelligence-powered autonomous experimentation (AI/AE) to condense 20-25 years of R&D into five.
Handwerker mentioned the Materials Genome Initiative, a multi-agency project dedicated to “discovering, manufacturing, and deploying advanced materials twice as fast and at a fraction of the cost compared to traditional methods.” It’s a joint effort with the Department of Defense, the Department of Energy, the National Institute of Standards and Technology (NIST), the National Institute of Health (NIH), the National Science Foundation (NSF), and the U.S. Geological Survey. The MGI scope covers a broad range of structural and functional materials, including metals, ceramics, and semiconductors.
Collaborating to Move Forward
There are some exciting possibilities for AI to accelerate new materials development in our industry with an overlaid sustainability lens, but automation can’t do all the work. Scientists and engineers still need to synthesize and test new materials so they can go from the lab to the fab. That piece requires intense collaboration between suppliers and customers and between universities and businesses.
Companies that want to leverage AI to develop more sustainable materials and processes might want to get involved on the industry level. SEMI’s Smart Manufacturing Initiative is building a roadmap incorporating best practices for reducing GHG emissions, energy, and water consumption. The next topic they are tackling is chemical waste reduction. AI-based tools are part of the roadmap. The more companies participate, the better the roadmap can reflect current and future needs.
Like any technology, AI can solve problems and create new ones. Here’s hoping that our industry makes the best use of AI to develop safer, more effective, and more
